These are chat archives for ipython/ipython

5th
Mar 2016
Jessica B. Hamrick
@jhamrick
Mar 05 2016 20:10 UTC
I am all of a sudden seeing a bunch of failing test cases on travis in nbgrader with "Kernel died before replying to kernel_info"
Has anybody else seen this on other projects?
I’ve been pretty much just working on documentation, so I don’t see how anything I’ve changed could be producing these failures
Jessica B. Hamrick
@jhamrick
Mar 05 2016 20:32 UTC
Looks like @parente is getting the same error on builds for the kernel gateway
Carol Willing
@willingc
Mar 05 2016 21:06 UTC
@jhamrick It’s not you. It’s Travis and/or a legitimate issue in our code base. I’ve been working on resolving for @parente on kernel gateway. Travis was down for maintenance about an hour ago. I’ll rekick it on nbgrader and see what happens.
Carol Willing
@willingc
Mar 05 2016 22:01 UTC
@jhamrick I’ve rebuilt some of failures on Travis and the failures appear to be related to timeouts vs. nbgrader code.
Jessica B. Hamrick
@jhamrick
Mar 05 2016 22:18 UTC
@willingc Hmm, weird
Carol Willing
@willingc
Mar 05 2016 22:18 UTC
@jhamrick Tell me about it.
Jessica B. Hamrick
@jhamrick
Mar 05 2016 22:18 UTC
So travis is timing out? Or the kernel is?
Looks like nbgrader is doing better at least (fingers crossed)
Carol Willing
@willingc
Mar 05 2016 22:21 UTC
Not sure. It may be a low level websocket issue. More likely, it is Travis throttling under load and then something in the websocket times out which gets propagated up. I’ve been babysitting the nbgrader pr and rerunning the individual fails.
Jessica B. Hamrick
@jhamrick
Mar 05 2016 22:22 UTC
Hmm
Carol Willing
@willingc
Mar 05 2016 22:22 UTC
@jhamrick Out of curiosity. When you run nbgrader tests locally, how long does it take?
Jessica B. Hamrick
@jhamrick
Mar 05 2016 22:23 UTC
Ok, well no need to babysit it unless it’s helpful for figuring out what’s going wrong… it doesn’t seem like restarting always fixes it. I tried to restart the builds a whole bunch earlier today and they kept failing so I gave up
Umm, I don’t know exactly; I usually only run the tests that are relevant to the part I’m working on
Let me run the full test suite now and I’ll let you know when it’s done :-)
Carol Willing
@willingc
Mar 05 2016 22:26 UTC
Actually, the individual tests that fail are all different. I’m babysitting to see if there is anything different than what I was seeing on kernel gateway where I had the test verbosity set high. Thanks for running the full suite.
Jessica B. Hamrick
@jhamrick
Mar 05 2016 22:27 UTC
docs on python3.5 took 1.6 minutes
Carol Willing
@willingc
Mar 05 2016 22:29 UTC
what about the python tests
They are taking close to 9ish minutes on Travis
Jessica B. Hamrick
@jhamrick
Mar 05 2016 22:35 UTC
formgrader on python 3.5 took 7.5 minutes
I will run python next
Carol Willing
@willingc
Mar 05 2016 22:35 UTC
Cool thanks!
Jessica B. Hamrick
@jhamrick
Mar 05 2016 22:44 UTC
python on python 3.5 took 8 minutes
nbextensions on python 3.5 took 3 minutes
I wish I could speed up the python tests
Carol Willing
@willingc
Mar 05 2016 22:54 UTC
File an issue on the test speed. I think I know someone at PyCon that can help with that at the end of May. I did limit the number of concurrent Travis jobs to 1 while I troubleshoot. I will disable back to the original Travis setting after I’m done.
Jessica B. Hamrick
@jhamrick
Mar 05 2016 22:55 UTC
Ok
Carol Willing
@willingc
Mar 05 2016 22:55 UTC
Thanks :D
Jessica B. Hamrick
@jhamrick
Mar 05 2016 22:55 UTC
I think the issue with speed is related to a more general issue with the jupyter application and traitlets being slow
But it would be nice if I could make it not quite so painfully slow (it takes about 1s just to run the command nbgrader with no other arguments)
Carol Willing
@willingc
Mar 05 2016 22:56 UTC
That may be true. I know some core devs on py.test and Twisted that might be willing to take a peak at PyCon.
Jessica B. Hamrick
@jhamrick
Mar 05 2016 22:57 UTC
Sweet :-)
I may try poking at it myself a bit before then but chances are I won’t get very far with it, so that would be awesome
Carol Willing
@willingc
Mar 05 2016 22:58 UTC
+1